An Improved Subspace Identification of a Nonlinear Soil- Structure System

In this paper, we analyze experimental results from a centrifuge model test simulating seismic response of a pile foundation in soft clay with a cement-treated improvement zone. We apply a robust frequency domain nonlinear subspace identification method known as FNSI to displacement data derived from strain gages along the pile and from a linear potentiometer attached to the superstructure. The earthquake base motion is considered to be a strong excitation capable of introducing nonlinearities in the soil-pile-superstructure system represented by a 2-DOF mechanical model with a coupling spring between the superstructure mass and the foundation mass as well as a ground spring between the foundation mass and the base. Nonlinearities introduced by soil stiffness degradation depending on soil strain level are lumped into the ground spring. Inclusion of a softening cubic spring in the nonlinear model formulation proved to be highly beneficial. The identified modal parameters are compared with those obtained from the time domain linear subspace state space system identification method. Our results demonstrate the power of the FNSI method and show significant improvement in the predictions of the identified nonlinear state-space model despite conditioning issues with the extended input information matrix.